Overview
- Community reports cite a misconfigured Vertex AI log showing the identifier claude-sonnet-5@20260203 with an internal codename “Fennec” and TPU optimization flags.
- The date-stamped identifier aligns with Anthropic’s model versioning format used for prior releases, which lends plausibility to the reference.
- Anthropic has issued no announcement or documentation on the model, and the earliest public trace is a Twitter screenshot that lacks proof of direct log access.
- Unconfirmed claims circulating in reports include 80.9%+ SWE-Bench results, very large context options, a “Dev Team” sub-agent mode, and lower pricing than Opus 4.5.
- Dataconomy notes TestingCatalog’s brief hands-on with a non-thinking build reporting strong math and coding in some workflows and a 128k context window, with final specifications unspecified.